Prediction of bioconcentration factors (BCFs) and bioaccumulation factors (BAFs) for per- and polyfluoroalkyl substances (PFASs) using Read-Across and q-RASPR

Abstract

Per- and polyfluoroalkyl substances (PFASs) contamination poses an environmental concern due to their ability to bioaccumulate in aquatic species and adversely impact human health. Experimental bioconcentration factor (log BCF) data of freshwater fish (Teleostei taxonomic class) for representative PFASs were used to develop the quantitative structure-property relationship (QSPR) and machine learning (ML)-based quantitative Read-Across Structure-Property Relationship (q-RASPR) models. We utilized various ML algorithms to effectively consider both linear and non-linear relationships. External predictions from the best-performing ML q-RASPR model (Q²_F1 = 0.930, Q²_F2 = 0.917, MAE_test = 0.491, RMSE_test = 0.653) were better than the corresponding QSPR model and a previously reported model. In compliance with the Registration, Evaluation, Authorization, and Restriction of Chemicals (REACH) guideline, a true external set prediction of 2411 unknown PFASs was performed, and they were classified into bioaccumulation categories as per Annex XIII. The bioaccumulation factor (log BAF) of PFASs was predicted using the Read-Across approach, and the predictivity and reliability of the method were assessed. Additionally, we have developed a Python-based tool PFAS_(BCF)_Predictor-v1.0 (available from https://sites.google.com/jadavpuruniversity.in/dtc-lab-software/home/pfas_bcf_predictor) to predict the BCF value of a true external set and classify PFASs into bioaccumulation categories according to the REACH guideline (Annex XIII), thus emphasizing the overall applicability and interpretability of this study. Statistical analysis suggests that the bioconcentration factor of PFASs depends on the number of CF₂ groups, the chain length of the molecule, and the atomic distribution properties. The developed models will further assist in designing an environmentally conscious strategy and control measures for PFAS contamination.

Keywords: BAF; BCF; Machine learning; PFASs; QSPR; REACH; q-RASPR.